How preferences flowed in South Australia

We have almost reached the end of the House of Assembly count for the recent South Australian state election. We know the winner in all 47 seats. In 46 seats, we have a full distribution of preferences. The only exception is in Narungga, where a recount confirmed that One Nation won the seat by a 77-vote margin over the Liberal Party. We don’t yet have a full distribution of preferences for Narungga, and I assume we’ll need to wait until after the Easter long weekend for that to be published.

UPDATE: I’ve now added the Narungga data to this blog post. The rest of this blog post reflects all 47 seats.

Pretty much all of the analysis I am doing for this blog post is based on the three-candidate-preferred count, which is the second-last round of the preference count.

This South Australian election was extremely complex, with doubt about who would come in the top two, and with preferences potentially playing a larger role than in the past. So for this post I will be going through what the final preference distributions tell us about the shape of South Australia’s election: how preferences flowed, how close the gap was between second- and third-placed candidates, and how many seats featured a particular combination of candidates.

For my next blog post, I am going to use this data to publish not just the actual margins, but also my estimates of alternative margins, where it was plausible that a different candidate could have made the top two.

Let’s start by looking at how often a particular party made the three-candidate-preferred count.

Until 2026, Labor and Liberal made the top three in almost every seat. Over the previous six elections, Labor missed the top three twice, and the Liberal Party never missed the top three.

This year, Labor missed out in four seats, and the Liberal Party missed out in sixteen! One Nation reached the top three in forty seats. The Greens also had a significant downturn in the number of seats where they made the 3CP.

From 2002-2022 (and likely for much longer) there was always a third spot in the 3CP and that position has varied. The Australian Democrats had that role in most seats in 2002, and SA Best dominated in 2018. Otherwise, the Greens have usually made the count in a majority of seats, while Family First regularly made the 3CP in 2006-2014.

The specific combination of 3CP sets is:

• ALP-LIB-ON – 20
• ALP-ON-GRN – 13
• ALP-LIB-GRN – 6
• LIB-ON-IND – 4
• ALP-ON-IND – 3
• ALP-LIB-IND – 1

And the specific combination of 2CP sets is:

• ALP-ON – 25
• ALP-LIB – 13
• LIB-ON – 4
• ON-IND – 2
• ALP-GRN – 1
• ALP-IND – 1
• LIB-IND – 1

The next question I was wondering about was whether the gap between second and third has been shrinking. We have noticed this trend in federal politics, where we have more seats where the 3CP count becomes critical. It turns out that the gap has shrunk slightly, but is roughly back to where it was in 2018 when SA Best made a splash.

This actually rings true when considering the actual results. While there was a lot of seats where the top two wasn’t known before election day, there were only a few where it was unclear after election day, and most of those became clear without needing a full 3CP count.

Four seats had a gap of less than 4% between second and third on the 3CP:

• Heysen – ALP 0.4% over GRN
• Croydon – GRN 0.6% over ON
• Schubert – ALP 2.2% over ON
• Enfield – ON 2.9% over LIB

Only in Heysen did this come close to mattering for the result, but I’ll come back to that in the next post.

So did preferences make much of a difference to the result?

There were only two seats where the winner came from behind. In the other 45 seats, the primary vote leader ended up winning. In Finniss, independent Lou Nicholson came from fourth place to win, overtaking the Liberal primary vote leader. In Kavel, independent Matt Schultz overtook Labor.

In the seat of Hammond, One Nation topped the primary vote but actually fell behind on the three-candidate-preferred vote, into second place, and relied on preferences from the Liberal candidate to defeat Labor.

So overall it doesn’t look like preferences had a big impact one way or another, but there were a lot of them. Now that we have 3CP data, we can calculate how preferences flowed on the final round of the count. This is not the same as having preference data based on the candidate’s primary vote (apparently we will get this at the end of the year) but it does have the benefit of isolating other effects.

When comparing primary vote figures to the final preference count, it is difficult to isolate preferences from particular partisan elements – for example, how Liberal preferences flow. But by taking the second-last round of the count, there’s just the votes of one party left to be distributed. This does include votes that were cast for another candidate, but that’s not a bad thing.

So this next chart shows the proportion of preferences that flowed from each party for each two-candidate-preferred pairing. I’ve noted how many seats had this preference flow.

The most common 2CP contest was Labor vs One Nation. About half had the Greens in third, and most of the rest had the Liberals in third. Greens preferences flowed to Labor at about 80%, while Liberal preferences flowed at two thirds to One Nation.

In the Labor-Liberal contests, there is a similar mirror image, but even more polarised – Greens preferences were even stronger to Labor, while One Nation preferences favoured Liberal even more. But again, the parties of the left have tighter preference flows amongst each other than the parties of the right.

When you break down this data by electorate, there are some interesting trends. Greens preferences tend to be a little more pro-Labor in a race against the Liberal Party. In the seat of Elizabeth, almost a third of Greens preferences flowed to One Nation. That was a seat where a donkey vote would go to the Greens then One Nation, and it is a relatively weak area for the Greens.

Liberal preferences favoured One Nation over Labor in every seat, ranging from 53% in Enfield to 73.5% in Hammond. In general, One Nation does not favour Liberal quite so strongly.

Preferences do clearly have a skew, but it seems that multi-party politics means that they often cancel each other out. So I thought I would look at how much preferences helped each party, subtracting the party’s primary vote from their 2CP vote in every seat where they made the 2CP.

The political parties all benefited to similar degrees, but unsurprisingly One Nation does worst out of preferences, with Liberal doing a bit worse than Labor. The Greens didn’t do that well, but this only covers one seat (Croydon) which isn’t exactly Greens heartland.

It is remarkable, however, how well independents who made the final preference count benefited.

This can be seen clearly in my final chart, which shows the preference gain for each candidate in the top two, compared to that candidate’s primary vote.

Labor candidates fairly consistently did better than One Nation and Liberal candidates, but three independent candidates are way off the chart. Independents in Finniss, Kavel and Mount Gambier gained 30-40 percentage points from preferences. No major party candidate gains more than 28%.

That’s it for this post – in the next one, let’s look at some ways to understand the marginality of seats as we lose the simplicity of the pendulum.

Liked it? Take a second to support the Tally Room on Patreon!